An investigation of the factors affecting medical expenses: The case of Taiwan’s NHI implementation

Journal of Applied Finance & Banking, vol. 6, no. 6, 2016, 157-166
ISSN: 1792-6580 (print version), 1792-6599 (online)
Scienpress Ltd, 2016

An investigation of the factors affecting medical
expenses:
The case of Taiwan’s NHI implementation
Tsung-Yu Hsieh1, Huai-I Lee2 and Yu-Ju Huang3

Abstract
This paper uses annual data for Taiwan from 1970 to 2003 to examine the
factors affecting medical expenses. The results are as follows: (1) after the
implementation of the National Health Insurance system, the average per capita
health care spending increased significantly, by about 16%. (2) The income
elasticity of health care spending is greater than 1, which means that it is a luxury
good. This is in contrast with the findings of other Taiwanese studies, but supports
the results in the foreign literature. (3) Increasing the number of physicians may
cause “supply-induced demand”, but other explanatory variables may also affect
this. (4) Taiwan has an aging population, and the proportion of ageing population
has a positive correlation with health care expenditure. (5) Universal health
insurance might lead to an ex post moral hazard.
JEL classification numbers: I13, I11, D82
Keywords: moral hazard, health care spending, supply-induced demand

1 Introduction
A universal National Health Insurance (NHI) system has been in place in
Taiwan since March 1995. However, due to the rapid ageing of the population,
coupled with advances in medical technology, the demand for medical care is
rising, along with the related costs. All other things being equal, health insurance
income will eventually be lower than the related health insurance spending, and
1

Department of Finance, Ming Dao University, Taiwan
Department of Marketing and Distribution Management, WuFeng University, Taiwan
3
Nurse Department, National Cheng Kung University Hospital Douliou Branch, Taiwan
2

Article Info: Received : August 18, 2016. Revised : August 31, 2016.
Published online : November 1, 2016


158

Tsung-Yu Hsieh et al.

the health insurance system will thus face a financial crisis. It is thus necessary to
address the issue of health care financing in this context.
Taiwan’s NHI can reduce the cost of medical treatment for patients, and this
may lead to a so-called “moral hazard”, as people may have little incentive to
reduce the amount of resources they use (Pauly, 1968; Leu, 1986). This article will
thus explore whether the implementation of NHI has affected health care
expenditures in Taiwan.
Using the data for the period 1970 to 2003, the following issues are examined
in this work. 1. whether health care spending is a necessity or a luxury using the
income elasticity of health care spending. 2. Whether an increase in the number of
physicians produces induced demand. 3. Whether there is a positive correlation
between the proportion of ageing population in Taiwan and health care spending.
4. Whether the popularity of the NHI system has led to a moral hazard. 5. Whether
the relationship between health care and health care spending varies with the
implementation of the NHI system.

The rest of this paper is divided into five parts, as follows. Section 1
introduces this study, followed by the literature review in Section 2. Section 3
presents the methodology used in this work, then Section 4 pres4tnes the results of
the empirical research. Section 5 then ends this paper with a conclusion.

2 Literature review
2.1. Cross-sectional study
In examining 13 OECD countries in similar levels of economic development,
Newhouse (1977) discusses the relationship between real income (as measured by
gross domestic product) and the medical expenses. The empirical findings indicate
that the R-square is almost 90% in all the regression analysis examined, and the
elasticity of medical expenses is larger than one, and thus health care is a luxury
good. This study was the first to examine what factors affect health care spending,
an issue that has been taken up by many subsequent scholars. “Non-institutional”
factors have been shown to play a large role in explaining medical expenses(Leu,
1986). Newhouse (1977) also reports similar findings. After adding several other
factors, Gerdtham et al (1992) report the following results. First, national income
per capita, level of urbanization, degree of government intervention and the
capitation payment all have a positive impact on real income, but realincome is
still the most important factor. Next, health care is a luxury good, and in general
the effects of government intervention on real income are not high. Finally, the
high average amount of unpaid medical expenses leads to the high average health
care expenditure.
These earlier works were all cross-sectional comparative studies among
different countries, and they did not consider exchange rates and relative


An investigation of the factors affecting medical expenses …

159

purchasing power. Therefore, some biases may exist in their results. After
considering exchange rates, Parkin et al. (1987), Gerdtham and Josson (1991), and
Hitiris and Posnett (1992) find that the main source of real income is medical
expensesand medical expenses are a luxury good, and that non-real income
variables have a very small influence on medical expenses.
2.2 Time series analysis
In recent years, many scholars in various countries have examined the main
factors affecting health care spending using time series analysis. Using data for 22
OECD countries from 1970 to 1991, Gerdtham et al. (1998) report the following
results: (1) There is a positive correlation among real income, cigarette
consumption and health care expenditure; and (2) health care expenditure is a
necessity rather than luxury good. However, these results may be due to the nonstationary characteristics among these variables, leading to spurious regressions.
Therefore, after considering both co-integration and an error correction model,
Murthy and Ukpolo (1994) examine US data over the period 1960~1987, and
obtain the following results: (1) Health care is a necessary good, and not a luxury.
(2) Population aging and health care expenditure are positively correlated. (3) The
number of physicians is positively associated with health care spending (Hansen
& King, 1996). Also using co-integration analysis, Gerdtham and Lothgren (2000,
2002) find a long-term relationship between the health care spending and income.
Using Swiss data and regression analysis, Zweifel et al. (1999) find that average
life expectancy is a more important factor than population aging with regard to
health care spending. Using US data for the period 1982~1990, Newhouse (1992)
finds that changes in the effects of both demand- and supply-side variables were
caused by technological innovations during this time.4

3 Methodology
3.1. Data sources and variable estimation
Based on the previous literature, this study assumes that the following variables
have significant effects on per capita health care expenditure: (1) The demand

side: the degree of population aging and level of real income. (2) The supply side:
the provision of universal health insurance and the number of physicians
compared to the total population. Annual data from 1970 to 2003 is used in this
study, and all variables are scaled by the natural logarithms. The variables are
defined below, and the related data sources are given:
3.1.1 The average health care expenditure per person (HCE)
The NHI system in Taiwan has been adopted since 1996, and the main source of
4

The main variables are the degree of population aging, real income, universal health insurance,
physician population, and the relative price of medical services.


160

Tsung-Yu Hsieh et al.

data for this is the “Republic of China Statistical Yearbook”, as published by the
Directorate General of Budget, Accounting and Statistics.
3.1.2 The demand side
a. Degree of population aging (OLD)
Grossman (1972) states that the health depreciation rate of individuals will
accelerate as they are ageing, and thus the need for medical treatment will
increase. Medical expenses should thus increase with the degree of population
ageing. The data used to assess this is taken from the “Republic of China
Statistical Yearbook”, as noted above. The degree of population ageing is defined
as the population aged 65 or older divided by the total population in a given year.
b. Real income (GDP)
As national income grows the demand for health care will increase, and this will
then raise health care expenditures. The data to measure this are taken from the

Ministry of Education AREMOS economic statistics database, “Taiwan Area
National Income Repository.” Real income is defined as the ratio of GNP divided
by the population of this country in a given year.
3.2 The supply side
3.2.1. Universal health insurance (INSURE)
The provision of universal health insurance reduces the direct medical costs,
further giving individuals little incentive to reduce health care spending. The data
on universal health insurances is taken from the following databases: (1) The
number of men in employment is obtained from the Central Trust of “Public
Servants’ Insurance Statistics” and the Bureau of Labor’s “Taiwan and Fujian Area
Labor Insurance Statistics”. (2) Details of the implementation of the National
Health Insurance system are taken from the statistics published by the Central
Bureau of National Health Insurance. This is measured as follows:
 The total number of agricultural and fishery workers participated in public insurance
; Before the implementation of the NHI

The total population for the year

Universal health insurance  
The total number of health care participated in the referendum

; After the implementation of the NHI

The total population for the year

(1)

3.2.2 Health insurance implementation (I)
After the implementation of after-service health insurance the popularity of such
insurance has increased, and this may also increase the related moral hazard,

further rasing health care spending. As the universal health insurance system in
Taiwan started on 1995/03, the year 1995 will be used as a cutoff point. I is 1
before 1995, and 0 otherwise.


An investigation of the factors affecting medical expenses …

161

3.2.3 Number of doctors per 10,000 persons (Doctor)
The number of doctors per 10,000 individuals is defined as ratio, the total
number of doctors divided by the total population in Taiwan multiplied by 10,000.

4 The empirical results
4.1. The basic descriptive statistics
Table 1 shows that the individual data are generally subject to the assumption of
normality. The average per capita health care spending was about four times
higher before the implementation of NHI (20,320.37 / 4,944.444) than after it, the
proportion of the population ageing was about twice as high after the
implementation than before(0.0831/0.04674). The average per capita income in
real terms was almost three times higher before the implementation of NHI
(401,473.9341/166,879.74) than after. The insurance coverage ratio great four-fold
after the implementation of NHI (0.9608/0.2244), while the average number of
physicians per capita doubled.
The average per capita income in real terms increased from 64,824 in 1970
to 444,842 in 2003, a seven-fold rise in 33 years, while the proportion of the
population ageing increased by about three-fold (0.091095/0.029706). The
average number of doctors per capita increased four times (0.002054/0.0005), and
the sodium retention ratio also increased from 10% to 98%. The average growth
rate of per capita health expenditure rose 17-fold over the 33 years from 1970 to

2003. With increases in the quality of life and personal income, people will pay
more attention to their health, and thus health-related spending increases. Of
course, both technological improvements and growth in the ageing population can
explain the rise in health care spending.The statistics reported above indicate that
after the implementation of NHI in Taiwan health care expenses grew much faster
than the economy in in real terms. We can thus see that growth and income cannot
fully explain the majority of health care spending, in part because of the
population ageing in this period (with the elderly accounting for about 10% of the
total population), and a substantial increase in insurance rates The growth in the
number of doctors also may explain the rise in medical spending, via so-called
“the doctor supply-induced demand”. However, advances in medical science and
technology also play important roles in the rise in spending.


Tsung-Yu Hsieh et al.

162

Table 1: The descriptive statistics
Variable

Mean

Median

Stdev.

Skewness

Kurtosis

Jarque-Bera Test

HCE

4944.44

3840.04

3343.56

1.17

3.26

5.78*

OLD

0.046

0.044

0.012

0.47

2.03

1.88

GDP

166879.74

143923.68

78311.65

0.53

2.06

2.09

INSURE

0.22

0.18

0.11

0.49

1.81

2.47

DOCTOR

0.0009

0.00091

0.0003

0.41

2.06

1.64

HCE

20320.37

21117.09

3151.37

0.39

1.83

0.74

OLD

0.08

0.08

0.005

0.008

1.87

0.47

GDP

401473.93

411244.88

35680.49

-0.46

1.92

0.76

INSURE

0.96

0.96

0.01

-1.57

5.14

5.44*

DOCTOR

0.0018

0.0018

0.0001

0.04

1.6

0.73

HCE

9014.54

4998.91

7612.37

0.81

2.14

4.78

OLD

0.056

0.052

0.019

0.3

1.69

2.92

GDP

228978.2

202183

125718.3

0.36

1.72

3.1

INSURE

0.41

0.28

0.34

0.81

1.95

5.29*

DOCTOR

0.0012

0.0011

0.0004

0.21

1.72

2.56

Panel A: Before 1995

Panel B: After 1995

Panel C: Overall period

Note: *denotes significant at the 10% level

4.2 Multi-collinearity analysis
The results of the KPSS test (which includes both intercept and trend terms),
show that all the variables are significant at the 0.05 level with the exception of
ln(GDP). The results are generally consistent with the basic assumptions of the
traditional least squares method, but may need to be evaluated to detect their
residuals in the following regression equation. A high degree of col-linearity
between these explanatory variables may lead to the biased empirical results, the
multicollinearity between these variables are needed to be examined. Table 2
shows significant multicollinearity among these variables (all the correlation
coefficient values are above 0.97), so only the simple regression analysis are
examined in the following regression analysis.


An investigation of the factors affecting medical expenses …

163

Table 2: Multicollinearity analysis
Variable

Mean

Median

Stdev.

Skewness

Kurtosis

Jarque-Bera Test

HCE

4944.44

3840.04

3343.56

1.17

3.26

5.78*

OLD

0.046

0.044

0.012

0.47

2.03

1.88

GDP

166879.74

143923.68

78311.65

0.53

2.06

2.09

INSURE

0.22

0.18

0.11

0.49

1.81

2.47

DOCTOR

0.0009

0.00091

0.0003

0.41

2.06

1.64

HCE

20320.37

21117.09

3151.37

0.39

1.83

0.74

OLD

0.08

0.08

0.005

0.008

1.87

0.47

GDP

401473.93

411244.88

35680.49

-0.46

1.92

0.76

INSURE

0.96

0.96

0.01

-1.57

5.14

5.44*

DOCTOR

0.0018

0.0018

0.0001

0.04

1.6

0.73

HCE

9014.54

4998.91

7612.37

0.81

2.14

4.78

OLD

0.056

0.052

0.019

0.3

1.69

2.92

GDP

228978.2

202183

125718.3

0.36

1.72

3.1

INSURE

0.41

0.28

0.34

0.81

1.95

5.29*

DOCTOR

0.0012

0.0011

0.0004

0.21

1.72

2.56

Panel A: Before 1995

Panel B: After 1995

Panel C: Overall period

Note: *denotes significant at the 10% level

4.2. The regression analysis5
The ARCH test is executed for the regression equation to examine whether
the heterogeneity of variance is an issue. The empirical results show that ARMA
(1,1) -GARCH (1,1) model is an appropriate form in estimating regression
equations.
Based on Eq.(1), there was a significant increase in health care spending after
the implementation of NHI (per capita health care spending increased on average
16.826%). The low cost of medical treatment for individuals may lead to an ex
post moral hazard. Of course, this may also be due to technological or medical

advances.
The results of Eq.(2) indicate that population ageing was related to a significant
5

Because of the multicollinearity among variables exists, not all the variables are included in these explanatory variables
simultaneously to avoid problems in estimating the coefficients.


Tsung-Yu Hsieh et al.

164

increase in health care spending, although the rate of increase was less than the
increase in medical expenses. When the population ageing increases by 1%, then
medical expenses per capita will increase by 0.6%. As Taiwan now has an ageing
population, the government should carefully consider how to provide such
individuals with appropriate medical care while avoiding significant substantial
increases in spending.
The results of Eq.(3) show that the rate of increase in per capita medical
expenses was much larger than the increase in personal income in real terms, with
income increasing by 1%, and the average per capita health care spending rising
by about 1.3%. The health care is a luxury good, in contrast to earlier Taiwanese
studies but in line with much research carried out in other countries. Health care
spending will increase as the income increases.
The results of Eq.(4) show that medical insurance expenses increased
significantly, although the rate of increase was less than that seen for average
personal medical expenses. An increase of 1% in insurance coverage ratio led to
about a 0.3% increase in the average health care spending. One possible
explanation for this is that the expansion of the insurance system for medical
services has produced a rise in prices and quality over the last years, resulting in a

significant increase in medical expenses.
Table 3. The regression analysis


The regression equation: ln(HCEy )  ˆ0  ˆ1I y  ˆ2 ln(OLDy )  ˆ3 ln(GDPy )  ˆ4 ln(INSURE)  ˆ5 ln(DOCTOR)
Eq.

Intercept

I

(1) 14.47**

0.16**

(2) 13.86**

0.05**

(3) -7.64**

0.12**

(4)
(5) 3.83**

15.97 -0.14**
0.26**

Ln(OLD)

Ln(GDP)

Ln(INSURE)

Ln(DOCTOR)

R2

0.9976
0.65**

0.9976
1.33**

0.9958
0.28**

0.9971
0.98**

0.9937

Note: *denotes significant at the 10% level

The results of Eq.(5) show that the increase in the number of physicians per million
people was far below the average rate of increase in health spending per capita. As the
number of physicians increased by 1%, the average per capita health spending increased
by only 0.9%, the above results support the previous phenomena, doctor supply induced
demand. Large hospitals in Taiwan have recently purchased many innovative medical

devices, and all hospitals and clinics have seen a substantial increase in the number of
doctors. This has raised the issue of the waste of medical resources, which needs to be
addressed to solve the NHI’s financial problems.


An investigation of the factors affecting medical expenses …

165

5 Conclusion
This paper uses data for the period, 1970 to 2003 to explore the factors affecting
medical expenses in Taiwan. First, based on the KPSS test, no significant trends
among the five variables exist (with the exception of the average individual real
income), the results of this work are significantly different from those presented in
the previous literature. After adding the appropriate lags and the residual
heterogeneity correction, the following empirical results are obtained.
(1) After the implementation of NHI in Taiwan, the average per capita health
spending increased significantly (an increase of approximately 16%). (2) The
income elasticity of health care spending is greater than 1, and thus it is a luxury
good. (3) The increase in the number of physicians may have led to supplyinduced demand; (4) The increase in the proportion of the elderly population is
positively related to the amount health care spending. (5) Universal health
insurance might lead to an ex post moral hazard, but other variables are suggested
to be used. (6) The relationships among these variables and health care spending
in Taiwan still exist even after the system was introduced.
The relationship between rising medical expenses and the establishment of an
NHI system is encouraged to examine in the future research. In addition, since this
article uses annual data, future research could use data with a higher frequency,
such as monthly data, in order to examine the issues raised in this work.

References

[1] Grossman, M., On the Concept of Health Capital and the Demand for Health,
Journal of Political Economy, 1972, 80 ,2, 223-255.
[2] Gerdtham, Ulf-G., Søgaard, J., Andersson, F., and Jönsson, B., An
econometric analysis of health care expenditure: A cross-section study of the
OECD countries, Journal of Health Economics, 1992, 11, 63~84.
[3] Gerdtham, Ulf-G., and M. Lothgren, On Stationarity and Cointegration of
International Health Expenditure and GDP, Journal of Health Economics,
2000, 19 ,4, 461-475.
[4] Gerdtham, Ulf-G., and M. Lothgren, New Panel Results on Cointegration of
International Health Expenditure and GDP, Applied Economics, 2002, 34 ,
1679-1686.
[5] Hansen P., and A. King, Aggregate health care expenditure in the United
States: evidence from cointegration tests: a comment, Applied Economics
Letters, 1996,3, 37~39.
[6] Hitiris, T., and J. Posnett, The determinants and effects of health expenditure
in developed countries, Journal of Health Economics, 1992,11, 173-181.
[7] Leu, R.E., The Public-private Mix and International Health Care Costs., in
Culyer, A. J. and B. Jonsson eds, 1986, Public and Private Health Services:


166

[8]

[9]
[10]
[11]
[12]

[13]

Tsung-Yu Hsieh et al.

Complementaries and Conflicts, Oxford: Basic Blackwell.
Murthy, N. R. V,, and V. Ukpolo, Aggregate health care expenditure in the
United States: evidence from cointegration tests, Applied Economics, 1994,
26, 797~802.
Newhouse, J. P., Medical-care expenditure: a cross-national survey, Journal
of Human Resource, 1977,12, 115~125.
Newhouse, J. P., Medical Care Costs: How Much Welfare Loss?, Journal of
Perspective, 1992, 6, 3, 3-21.
Pauly, M. V., The Economics of Moral Hazard, American Economic Review,
1968, 58-61, 531-537.
Parkin , D., A. McGuire, and B.Yule, Aggregate health care expenditures and
national income-Is health care a luxury good?, Journal of Health Economics,
1987,6,109-127.
Zweifel, P., S. Felder, and M. Meiers, Ageing of Population and Health Care
Expenditure: A Red Herring?, Health Economics, 1999, 8, 485-496.